Arabic PDF to Word
Most PDF-to-Word converters destroy Arabic: words come out reversed, English terms and phone numbers get flipped, and formulas turn to noise. This one rebuilds the reading order from the page’s own geometry with the Unicode bidirectional algorithm, turns the PDF’s shaping glyphs back into ordinary editable Arabic letters, and hands Word a genuinely right-to-left document. Everything happens in your browser.
Why Arabic breaks in other converters, and what this does instead
1. The PDF stores Arabic backwards, on purpose
A PDF has no notion of reading order. It is a list of glyph-drawing instructions, and glyphs always advance left to right. The only way to render a right-to-left script is to draw the last letter of the sentence first, at the left of the line. A converter that simply concatenates what it extracts therefore gets the sentence backwards, and normally also gets it broken into a dozen pieces, because a single visual line is usually drawn as many separate chunks in whatever order suited the producer.
This tool ignores the order the chunks arrive in. It groups them into lines by their baseline, sorts each line by its x position - which is the page’s real visual order, independent of the producer - then runs a real (simplified) implementation of the Unicode Bidirectional Algorithm, UAX #9, backwards to recover the logical reading order. An English term, a phone number, a citation like “[12]” or an inline formula is a left-to-right island: it gets placed correctly inside the Arabic flow without being reversed itself.
2. The letters are not really Arabic letters
Arabic letters change shape depending on their neighbours, and a PDF stores the shaped form - a separate Unicode code point from the Arabic Presentation Forms blocks (U+FE70–U+FEFF and U+FB50–U+FDFF). Paste those straight into Word and the text looks approximately right but is not really Arabic: it will not re-join when you edit it, spell-check will not touch it, and search will not find it. Every one of those code points is mapped back to its base letter here, including the mandatory lam-alef ligatures (ﻻ ﻷ ﻹ ﻵ) which decompose into two letters.
3. Math is handled honestly, in two tiers
Simple inline expressions - the arithmetic and basic algebra that fills real Arabic textbook, homework and report PDFs, things like 5x + 3 = 18 or P = 2(l + w) - are reconstructed as ordinary editable text. That alone fixes the overwhelming majority of “my equations got scrambled” complaints, because what was actually broken was their placement inside the right-to-left line, not the symbols themselves. Superscripts and subscripts detected from baseline shifts become real Word superscript and subscript formatting.
Genuinely typeset mathematics - an integral, summation, radical or tall bracket, or an expression built up across three baselines - is not guessed at. That region is cropped out of the original page and embedded as a picture at exactly the right place in the paragraph, so it looks precisely as the author intended. The tool tells you how many regions were treated each way.
4. The Word file is really right-to-left
Right-aligning the text is not the same thing as making it right-to-left. Each Arabic paragraph gets a real <w:bidi/> direction, each Arabic run gets <w:rtl/>, and the font and size are set on the complex-script attributes Word actually uses for Arabic, not only the Latin ones. Left-to-right islands are written as their own runs without the RTL flag, so Word does not mirror their punctuation.
Questions fréquentes
Why does my Arabic PDF come out reversed or scrambled in other converters?
Because a PDF stores glyphs in the order they are drawn, not the order they are read, and for a right-to-left script those are opposite. Producers also split one visual line into many separate drawing operations, often in an arbitrary order. A converter that just concatenates what it extracts gets the words backwards and the pieces out of order. This tool throws that order away and rebuilds the line from the page geometry - baseline grouping plus x position - then applies the Unicode Bidirectional Algorithm (UAX #9) in reverse to recover the logical reading order.
What exactly comes out as editable text, and what comes out as an image?
Everything is editable text except regions that are genuinely two-dimensional typeset mathematics. A region is only turned into an image when it contains a structural symbol (an integral, summation, product, radical or a tall bracket built from glyph fragments) or when a cluster of expression-like runs carries both a superscript and a subscript, or spans three separate baselines - a built-up fraction. Ordinary inline arithmetic and algebra, a lone exponent like x squared, footnote markers, English terms, numbers and citations all stay as real text. The tool reports the count of each so you know exactly what happened.
Can you convert equations into editable Word equation objects?
No, and any tool that claims to do this reliably for arbitrary PDFs is overstating it. Recovering the semantic structure of an arbitrary typeset equation from a page of positioned glyphs is a research-grade problem. Rather than produce a confident-looking but wrong OMML equation, this tool keeps a complex equation as a pixel-accurate crop of the original page, placed inline where it belongs. That is a real limitation and it is stated plainly - but it is a far better outcome than the garbled line of symbols you get otherwise.
Does this work on a scanned Arabic PDF?
No. If the pages are images with no text layer, there is nothing to reconstruct - this tool rebuilds reading order from extracted text, it does not read pixels. The tool detects that case and tells you. Run the pages through OCR first with the Arabic Image to Text tool on this site, then bring the recognised text into Word. A converter that promises to fix a scanned Arabic PDF without OCR is not being straight with you.
Will the layout look exactly like the original PDF?
No - and that is the trade-off you are choosing when you ask for an editable document. Word reflows text rather than pinning it to page coordinates, so paragraphs, headings and equation images are emitted in reading order rather than at their exact original positions. Columns, text boxes, vector rules and colour backgrounds are not reconstructed. What you get is correct, editable, correctly-directional Arabic content. If you need a pixel-identical copy instead, use the Visual mode of the general PDF to Word tool, which puts each page in as a full-page picture.
What about the kashida, harakat and Persian or Urdu letters?
The kashida (tatweel, U+0640) is a purely typographic stretch with no meaning, so it is removed by default - you can turn that off. Harakat (short-vowel marks) are real content and are kept by default, with an option to strip them. Persian and Urdu letters written in presentation forms - peh, tcheh, jeh, gaf, keheh, farsi yeh, heh doachashmee and the rest - are mapped back to their base letters too, as are word ligatures like the ones for Allah and the honorifics.
Which font should I pick for the Word file?
Arial is the default because it ships with Word on both Windows and macOS and has complete Arabic coverage, so the document opens correctly for whoever you send it to. Times New Roman, Calibri and Tahoma are equally safe. Simplified Arabic and Traditional Arabic look more typographically Arabic but are not present on every machine, so a reader without them will see a substitute. Whichever you choose is written into the complex-script font attribute, which is the one Word actually applies to Arabic text.
Is my PDF uploaded to a server?
No. The PDF is read with pdf.js, the reading order is rebuilt, the equation crops are rendered on a canvas and the .docx is assembled with the docx library - all inside your browser. Nothing is sent anywhere, which also means confidential documents stay confidential.